OpenCode

Settings#

Add AI Tokens as a custom provider in opencode.json, in your project or in ~/.config/opencode/:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "aitokens/qwen3.8-27b",
  "provider": {
    "aitokens": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "AI Tokens",
      "options": {
        "baseURL": "https://api.aitokens.ch/v1",
        "apiKey": "{env:API_KEY}"
      },
      "models": {
        "qwen3.8-27b": {
          "name": "Qwen3.8 27B",
          "limit": { "context": 262144, "output": 8192 }
        },
        "qwen3-coder-30b": {
          "name": "Qwen3-Coder 30B",
          "limit": { "context": 32768, "output": 8192 }
        }
      }
    }
  }
}

The key can come from two places:

• {env:API_KEY}, as above, reads the environment variable API_KEY: export API_KEY="sk-at-…" before you start OpenCode;
• or leave apiKey out, run /connect in OpenCode, choose Other, type the provider ID aitokens and paste the key. OpenCode keeps it in ~/.local/share/opencode/auth.json.

limit.context is the context window and limit.output the most the model may write in one answer. Write them for every model: /v1/models does not publish them, and OpenCode has no other way to know them.

Which model#

qwen3.8-27b, for its 262,144 tokens of context; qwen3-coder-30b for quick tasks. Both returned tool calls on 11 October 2026, with and without streaming. More in Which model.

OpenCode talks to the API through the @ai-sdk/openai-compatible package named in npm. We tested that package with the same API on the Vercel AI SDK page.

Limits#

• output up to 8192. The API refuses a max_tokens above 8,192 with 422.
• No reasoning to show: the API does not return the reasoning of qwen3.8-27b.